xref: /openbmc/linux/arch/s390/mm/dump_pagetables.c (revision 09c434b8)
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/seq_file.h>
3 #include <linux/debugfs.h>
4 #include <linux/sched.h>
5 #include <linux/mm.h>
6 #include <linux/kasan.h>
7 #include <asm/kasan.h>
8 #include <asm/sections.h>
9 #include <asm/pgtable.h>
10 
11 static unsigned long max_addr;
12 
13 struct addr_marker {
14 	unsigned long start_address;
15 	const char *name;
16 };
17 
18 enum address_markers_idx {
19 	IDENTITY_NR = 0,
20 	KERNEL_START_NR,
21 	KERNEL_END_NR,
22 #ifdef CONFIG_KASAN
23 	KASAN_SHADOW_START_NR,
24 	KASAN_SHADOW_END_NR,
25 #endif
26 	VMEMMAP_NR,
27 	VMALLOC_NR,
28 	MODULES_NR,
29 };
30 
31 static struct addr_marker address_markers[] = {
32 	[IDENTITY_NR]		= {0, "Identity Mapping"},
33 	[KERNEL_START_NR]	= {(unsigned long)_stext, "Kernel Image Start"},
34 	[KERNEL_END_NR]		= {(unsigned long)_end, "Kernel Image End"},
35 #ifdef CONFIG_KASAN
36 	[KASAN_SHADOW_START_NR]	= {KASAN_SHADOW_START, "Kasan Shadow Start"},
37 	[KASAN_SHADOW_END_NR]	= {KASAN_SHADOW_END, "Kasan Shadow End"},
38 #endif
39 	[VMEMMAP_NR]		= {0, "vmemmap Area"},
40 	[VMALLOC_NR]		= {0, "vmalloc Area"},
41 	[MODULES_NR]		= {0, "Modules Area"},
42 	{ -1, NULL }
43 };
44 
45 struct pg_state {
46 	int level;
47 	unsigned int current_prot;
48 	unsigned long start_address;
49 	unsigned long current_address;
50 	const struct addr_marker *marker;
51 };
52 
53 static void print_prot(struct seq_file *m, unsigned int pr, int level)
54 {
55 	static const char * const level_name[] =
56 		{ "ASCE", "PGD", "PUD", "PMD", "PTE" };
57 
58 	seq_printf(m, "%s ", level_name[level]);
59 	if (pr & _PAGE_INVALID) {
60 		seq_printf(m, "I\n");
61 		return;
62 	}
63 	seq_puts(m, (pr & _PAGE_PROTECT) ? "RO " : "RW ");
64 	seq_puts(m, (pr & _PAGE_NOEXEC) ? "NX\n" : "X\n");
65 }
66 
67 static void note_page(struct seq_file *m, struct pg_state *st,
68 		     unsigned int new_prot, int level)
69 {
70 	static const char units[] = "KMGTPE";
71 	int width = sizeof(unsigned long) * 2;
72 	const char *unit = units;
73 	unsigned int prot, cur;
74 	unsigned long delta;
75 
76 	/*
77 	 * If we have a "break" in the series, we need to flush the state
78 	 * that we have now. "break" is either changing perms, levels or
79 	 * address space marker.
80 	 */
81 	prot = new_prot;
82 	cur = st->current_prot;
83 
84 	if (!st->level) {
85 		/* First entry */
86 		st->current_prot = new_prot;
87 		st->level = level;
88 		st->marker = address_markers;
89 		seq_printf(m, "---[ %s ]---\n", st->marker->name);
90 	} else if (prot != cur || level != st->level ||
91 		   st->current_address >= st->marker[1].start_address) {
92 		/* Print the actual finished series */
93 		seq_printf(m, "0x%0*lx-0x%0*lx ",
94 			   width, st->start_address,
95 			   width, st->current_address);
96 		delta = (st->current_address - st->start_address) >> 10;
97 		while (!(delta & 0x3ff) && unit[1]) {
98 			delta >>= 10;
99 			unit++;
100 		}
101 		seq_printf(m, "%9lu%c ", delta, *unit);
102 		print_prot(m, st->current_prot, st->level);
103 		while (st->current_address >= st->marker[1].start_address) {
104 			st->marker++;
105 			seq_printf(m, "---[ %s ]---\n", st->marker->name);
106 		}
107 		st->start_address = st->current_address;
108 		st->current_prot = new_prot;
109 		st->level = level;
110 	}
111 }
112 
113 #ifdef CONFIG_KASAN
114 static void note_kasan_early_shadow_page(struct seq_file *m,
115 						struct pg_state *st)
116 {
117 	unsigned int prot;
118 
119 	prot = pte_val(*kasan_early_shadow_pte) &
120 		(_PAGE_PROTECT | _PAGE_INVALID | _PAGE_NOEXEC);
121 	note_page(m, st, prot, 4);
122 }
123 #endif
124 
125 /*
126  * The actual page table walker functions. In order to keep the
127  * implementation of print_prot() short, we only check and pass
128  * _PAGE_INVALID and _PAGE_PROTECT flags to note_page() if a region,
129  * segment or page table entry is invalid or read-only.
130  * After all it's just a hint that the current level being walked
131  * contains an invalid or read-only entry.
132  */
133 static void walk_pte_level(struct seq_file *m, struct pg_state *st,
134 			   pmd_t *pmd, unsigned long addr)
135 {
136 	unsigned int prot;
137 	pte_t *pte;
138 	int i;
139 
140 	for (i = 0; i < PTRS_PER_PTE && addr < max_addr; i++) {
141 		st->current_address = addr;
142 		pte = pte_offset_kernel(pmd, addr);
143 		prot = pte_val(*pte) &
144 			(_PAGE_PROTECT | _PAGE_INVALID | _PAGE_NOEXEC);
145 		note_page(m, st, prot, 4);
146 		addr += PAGE_SIZE;
147 	}
148 }
149 
150 static void walk_pmd_level(struct seq_file *m, struct pg_state *st,
151 			   pud_t *pud, unsigned long addr)
152 {
153 	unsigned int prot;
154 	pmd_t *pmd;
155 	int i;
156 
157 #ifdef CONFIG_KASAN
158 	if ((pud_val(*pud) & PAGE_MASK) == __pa(kasan_early_shadow_pmd)) {
159 		note_kasan_early_shadow_page(m, st);
160 		return;
161 	}
162 #endif
163 
164 	for (i = 0; i < PTRS_PER_PMD && addr < max_addr; i++) {
165 		st->current_address = addr;
166 		pmd = pmd_offset(pud, addr);
167 		if (!pmd_none(*pmd)) {
168 			if (pmd_large(*pmd)) {
169 				prot = pmd_val(*pmd) &
170 					(_SEGMENT_ENTRY_PROTECT |
171 					 _SEGMENT_ENTRY_NOEXEC);
172 				note_page(m, st, prot, 3);
173 			} else
174 				walk_pte_level(m, st, pmd, addr);
175 		} else
176 			note_page(m, st, _PAGE_INVALID, 3);
177 		addr += PMD_SIZE;
178 	}
179 }
180 
181 static void walk_pud_level(struct seq_file *m, struct pg_state *st,
182 			   p4d_t *p4d, unsigned long addr)
183 {
184 	unsigned int prot;
185 	pud_t *pud;
186 	int i;
187 
188 #ifdef CONFIG_KASAN
189 	if ((p4d_val(*p4d) & PAGE_MASK) == __pa(kasan_early_shadow_pud)) {
190 		note_kasan_early_shadow_page(m, st);
191 		return;
192 	}
193 #endif
194 
195 	for (i = 0; i < PTRS_PER_PUD && addr < max_addr; i++) {
196 		st->current_address = addr;
197 		pud = pud_offset(p4d, addr);
198 		if (!pud_none(*pud))
199 			if (pud_large(*pud)) {
200 				prot = pud_val(*pud) &
201 					(_REGION_ENTRY_PROTECT |
202 					 _REGION_ENTRY_NOEXEC);
203 				note_page(m, st, prot, 2);
204 			} else
205 				walk_pmd_level(m, st, pud, addr);
206 		else
207 			note_page(m, st, _PAGE_INVALID, 2);
208 		addr += PUD_SIZE;
209 	}
210 }
211 
212 static void walk_p4d_level(struct seq_file *m, struct pg_state *st,
213 			   pgd_t *pgd, unsigned long addr)
214 {
215 	p4d_t *p4d;
216 	int i;
217 
218 #ifdef CONFIG_KASAN
219 	if ((pgd_val(*pgd) & PAGE_MASK) == __pa(kasan_early_shadow_p4d)) {
220 		note_kasan_early_shadow_page(m, st);
221 		return;
222 	}
223 #endif
224 
225 	for (i = 0; i < PTRS_PER_P4D && addr < max_addr; i++) {
226 		st->current_address = addr;
227 		p4d = p4d_offset(pgd, addr);
228 		if (!p4d_none(*p4d))
229 			walk_pud_level(m, st, p4d, addr);
230 		else
231 			note_page(m, st, _PAGE_INVALID, 2);
232 		addr += P4D_SIZE;
233 	}
234 }
235 
236 static void walk_pgd_level(struct seq_file *m)
237 {
238 	unsigned long addr = 0;
239 	struct pg_state st;
240 	pgd_t *pgd;
241 	int i;
242 
243 	memset(&st, 0, sizeof(st));
244 	for (i = 0; i < PTRS_PER_PGD && addr < max_addr; i++) {
245 		st.current_address = addr;
246 		pgd = pgd_offset_k(addr);
247 		if (!pgd_none(*pgd))
248 			walk_p4d_level(m, &st, pgd, addr);
249 		else
250 			note_page(m, &st, _PAGE_INVALID, 1);
251 		addr += PGDIR_SIZE;
252 		cond_resched();
253 	}
254 	/* Flush out the last page */
255 	st.current_address = max_addr;
256 	note_page(m, &st, 0, 0);
257 }
258 
259 static int ptdump_show(struct seq_file *m, void *v)
260 {
261 	walk_pgd_level(m);
262 	return 0;
263 }
264 
265 static int ptdump_open(struct inode *inode, struct file *filp)
266 {
267 	return single_open(filp, ptdump_show, NULL);
268 }
269 
270 static const struct file_operations ptdump_fops = {
271 	.open		= ptdump_open,
272 	.read		= seq_read,
273 	.llseek		= seq_lseek,
274 	.release	= single_release,
275 };
276 
277 static int pt_dump_init(void)
278 {
279 	/*
280 	 * Figure out the maximum virtual address being accessible with the
281 	 * kernel ASCE. We need this to keep the page table walker functions
282 	 * from accessing non-existent entries.
283 	 */
284 	max_addr = (S390_lowcore.kernel_asce & _REGION_ENTRY_TYPE_MASK) >> 2;
285 	max_addr = 1UL << (max_addr * 11 + 31);
286 	address_markers[MODULES_NR].start_address = MODULES_VADDR;
287 	address_markers[VMEMMAP_NR].start_address = (unsigned long) vmemmap;
288 	address_markers[VMALLOC_NR].start_address = VMALLOC_START;
289 	debugfs_create_file("kernel_page_tables", 0400, NULL, NULL, &ptdump_fops);
290 	return 0;
291 }
292 device_initcall(pt_dump_init);
293